search

tyshiwo / DRRN_CVPR17

Code for our CVPR'17 paper "Image Super-Resolution via Deep Recursive Residual Network"
226 stars 82 forks source link

The input dim? #7

Closed XiaotianM closed 6 years ago

XiaotianM commented 6 years ago

Hello, I find that you have splited the training Images into 31x31 in your paper. However, in your deploy file, you set the input dim as (1,1,130,130) ?

tyshiwo commented 6 years ago

Hi, for training, considering both training time and memory storage, we use 31*31 patch; for inference, since our network uses fully convolutional neural network structure, you can feed any size you want.